The Art of the Second Prompt: Where All the Real Skill Lives
First prompts get all the attention and deserve none. What separates good AI users is what they say after the first answer.
Watch someone good use a chatbot and the reveal is that their first prompt is usually mediocre. The skill is entirely in the second prompt: the one that reacts to what came back. This is the least discussed and most valuable skill in the whole category.
The weak second prompt is 'make it better,' which is a wish, not an instruction. The strong ones are diagnostic. 'The third paragraph contradicts the first, fix the third.' 'You hedged in the conclusion; pick a side and defend it.' 'This is 30 percent too long, cut examples, keep arguments.' Specificity about the output beats cleverness in the input every time.
The counterintuitive part: a bad first answer is not a failure, it is information. When a draft comes back wrong in an interesting way, it has shown you what the model understood about your request, which is usually what you actually failed to specify. I have learned more about writing briefs from wrong outputs than right ones.
There is a stopping discipline here too. Two or three iterations is where quality peaks in my experience; by the fourth, you are polishing the model's habits instead of your idea, and starting over with a sharper brief beats a fifth round. The people who iterate ten times are fighting their own first prompt.
None of this is prompt engineering in the mystical sense. It is the same skill as briefing a colleague: say what you want, look at what came back, and correct precisely. The tools just made the feedback loop take seconds instead of days, which is why the skill suddenly matters so much.